Puppet

Chris Price wrote:

The execute_posix method:

a) Should not be called during windows test runs, and

Ideally we want to test as much of the behaviour on Windows as we can, just as we want to test as much Windows behaviour on our Unix systems as we can. The less places a test runs the less likely we are to catch a problem early.

b) Needs to take more care not to inadvertently set environment variables in the main ruby process.

Yeah. That test is a bad test, since assuming that nothing bad can possibly happen inside the forked child when it gets run in the parent is … terrible.

This would expose the issue of, for example, the work-around for fork vs ActiveRecord that will start closing file descriptors too. Having to know that much is a real problem.

We should probably better not unit test this at all, or abstract the platform details to an easy to stub interface if we must, than to assume we can make fork and yield equivalent safely.

I’ve committed a fix that should get the tests passing again, but would be happy to do further work on this test to get rid of the “fork” stub if desired.